FinTextQA
收藏资源简介:
由HSBC Lab、香港科技大学和哈佛大学联合构建的FinTextQA数据集,专为金融领域长篇问答而设计。该数据集包含了1262条高质量、可追溯来源的问答对,涉及六种问题类别,平均文本长度达19.7千字,并经五轮人工筛选。数据源包括金融教科书及诸如香港金融管理局、欧盟、美国联邦储备系统等政府机构网站,确保了内容的相关性与权威性。相较于一般问答数据集,FinTextQA更侧重于长篇回答,要求模型生成段落级别的回答。该数据集为金融领域的长篇问答任务提供了一个高质量、挑战性的基准,助力深化金融概念理解及提升金融领域助手能力。
The FinTextQA dataset, jointly constructed by HSBC Lab, The Hong Kong University of Science and Technology, and Harvard University, is purpose-built for long-form financial question answering. It comprises 1,262 high-quality, source-traceable question-answer pairs across six question categories, with an average text length of 19.7 thousand characters, and has undergone five rounds of manual screening. Its data sources include financial textbooks and official websites of government institutions such as the Hong Kong Monetary Authority, the European Union, and the Federal Reserve System, ensuring the relevance and authority of the content. Unlike general question answering datasets, FinTextQA places greater emphasis on long-form responses, requiring models to generate paragraph-level answers. This dataset provides a high-quality, challenging benchmark for long-form question answering tasks in the financial domain, helping to deepen the understanding of financial concepts and enhance the capabilities of financial domain assistants.

- 1FinTextQA: A Dataset for Long-form Financial Question Answering · 2024年



